The year the crime data broke, and what it still hides
American Factbook Research
Figures as of
In 2021 the FBI retired the summary system American police agencies had reported crime through since 1930 and moved to incident-level reporting. Agencies transitioned at different times, and many reported nothing at all that year.
The national violent crime count fell from 1,273,163 offenses to 932,565. Almost none of that is a fall in crime. The share of the population living in a jurisdiction whose agency reported fell from 95.9% to 77.0% — from 319,035,026 residents to 258,198,281. The count did not measure less crime; it measured fewer places.
The count and the rate moved by different amounts, and the gap is the story
A crime rate on this site divides by the population the reporting agencies served, not by the population of the country. That is what makes the two figures come apart in 2021: the count lost a quarter of its agencies, while the rate divided what was left by the smaller population it actually described, and moved from 399.1 to 361.2 per 100,000 residents.
So the rate is the more comparable of the two — but it is not comparable in the way a reader expects. It describes the places that reported, and in 2021 those places were not a sample of the country. They were whichever agencies had finished moving to the new system.
What that looks like in one state
California reported on 99.9% of its population in 2020 and 26.2% in 2021. Florida fell to 46.9%, and its published violent crime rate fell from 383.6 to 203.9 per 100,000.
Pennsylvania is the case that shows why the direction cannot be assumed. Its coverage fell to 42.4%, and its rate went up, from 326.1 to 403.9. Losing agencies does not push a rate down. It changes which places the rate is about, and the places left are sometimes higher-crime and sometimes lower. That is the whole difficulty: the error has no direction, so it cannot be corrected for, only disclosed.
Reporting recovered. Six jurisdictions have not
Nationally, coverage is back: 96.4% in 2025 — above where it was before the transition. The number of jurisdictions reporting on less than the coverage threshold went from 7 in 2020 to 18 in 2021 and stands at 6 now.
Those six are not a footnote. Louisiana is the thinnest at 78.0%, which means roughly a fifth of the state is absent from every figure published about it — including the rankings on this site, where it appears with a rate built from the part that reported.
What this brief does not establish
It does not say whether crime rose or fell in 2021. The data cannot answer that, because the places measured in 2021 are not the places measured in 2020. It does not say why any agency stopped reporting, or whether the agencies that stopped differ from those that continued in any way other than that they stopped. Where a rate is published for a low-coverage jurisdiction here and elsewhere on this site, it is the rate among the people those agencies served, and that is all it is.
What this cannot show
- Reporting to the FBI is voluntary. A change in a published total can be a change in which agencies reported rather than a change in offenses, and in 2021 it mostly was.
- Rates here divide by the population the reporting agencies served, which is the denominator the FBI uses for its own rates. For a jurisdiction below full coverage the rate describes the part that reported, and the agencies that reported are not a random sample of the jurisdiction.
- Counts are offenses reported to law enforcement. They are not convictions, and they do not measure offenses that were never reported to police.
- This brief compares 2020, 2021 and 2025. It draws no conclusion about whether crime itself rose or fell across the reporting change, because the data cannot separate that from the change in who reported.
Sources
- FBI Crime Data Explorer API, Federal Bureau of Investigation, Criminal Justice Information Services Division. U.S. Government work, public domain (17 U.S.C. 105). Retrieved 2026-09-19.
Every figure above is the recorded result of a query checked into the repository alongside this text. Continuous integration re-runs those queries and refuses the change if a number here and a number there disagree. Read the methodology, or the datasets behind it.
Published by American Factbook LLC. An independent publication.